Re: ov-java.cc-tst fails on Lubuntu 14.04

[Adrian]

> error: [java] java.lang.ClassCastException: java.lang.Double cannot be 
> cast to java.lang.Float

Hi,

I'm not sure if the following observation concerning the java interface is 
related to the topic, but maybe.

When I compare results I get from calculations that are actually done by a 
java library (input vector passed to a java lib), then I notice that the 
results are numerically slightly different on different octave versions.

- Octave 3.6.1 (windows) and octave 3.8.1 (linux 64bit) give the same results. 
The result is actually the same when using java only (reading in data in the 
java class instead of octave), so I'm convinced it is correct.

- Octave 4.0.0 (both on windows 64bit and linux 64bit) give slightly different 
results. The first 5 digits are the same, so for my use it is precise enough, 
but still a bit weird.

When I investigate, it seems that in octave 4.0.0 vectors are passed to the 
java class with lower precision (maybe float).


Sincerely,
Adrian

(Tested on OpenSuse 13.1 64bit, with OpenJDK 1.7.0 64bit with
octave 4.0.0 x86_64 from repositoriy "science")